Freight transport prediction using electronic waybills and machine learning

A waybill is a document that accompanies the freight during transportation. The document contains essential information such as, origin and destination of the freight, involved actors, and the type of freight being transported. We believe, the information from a waybill, when presented in an electronic format, can be utilized for building knowledge about the freight movement. The knowledge may be helpful for decision makers, e.g., freight transport companies and public authorities. In this paper, the results from a study of a Swedish transport company are presented using order data from a customer ordering database, which is, to a larger extent, similar to the information present in paper waybills. We have used the order data for predicting the type of freight moving between a particular origin and destination. Additionally, we have evaluated a number of different machine learning algorithms based on their prediction performances. The evaluation was based on their weighted average true-positive and false-positive rate, weighted average area under the curve, and weighted average recall values. We conclude, from the results, that the data from a waybill, when available in an electronic format, can be used to improve knowledge about freight transport. Additionally, we conclude that among the algorithms IBk, SMO, and LMT, IBk performed better by predicting the highest number of classes with higher weighted average values for true-positive and false-positive, and recall.